A Comparative Analysis of Different CNN Architectures for Malware Classification
Anisha Mahato, Rana Majumdar, Swarup Kr Ghosh · 2024
The concept of digital privacy remains a contested and often misunderstood topic in the modern, internet-dominated era. Despite numerous advancements in cybersecurity, the pervasive threats to data security continue to erode trust in digital systems. Malware, a term encompassing a wide range of malicious software designed to disrupt operations, compromise systems, or steal sensitive information, represents a major challenge in this regard. Its widespread presence has led to the perception that digital privacy is increasingly unattainable, casting a shadow over cyberspace and intensifying the urgency for robust countermeasures. This study delves into the critical task of malware classification, leveraging the Malimg dataset, a benchmark resource widely used for malware research. To tackle this problem, four cutting-edge Convolutional Neural Network (CNN) architectures-LeNet-5, AlexNet, VGG-16 and ResNet-are employed to train and test the dataset, with the objective of identifying the architecture best suited for this classification task. The models are rigorously evaluated using four essential performance metrics: Accuracy, Precision, Recall, and F1-score, which collectively provide a comprehensive measure of their effectiveness. Among these architectures, ResNet emerges as the most effective, achieving a stellar accuracy of 99.78% and excelling across all other evaluation parameters. This highlights ResNet's superior capability in accurately identifying and classifying malware, thereby offering a promising solution to mitigate the growing threat posed by malicious software in cyberspace.